NeuroImage 229 (2021) 117752

Contents lists available at ScienceDirect

NeuroImage
journal homepage: www.elsevier.com/locate/neuroimage

Technical Note

Effect of wearing a face mask on fMRI BOLD contrast

Christine S.W. Law a,∗, Patricia S. Lan b,c, Gary H. Glover b,c
a
Systems Neuroscience and Pain Lab, Department of Anesthesiology, Perioperative and Pain Medicine, Stanford University School of Medicine, Palo Alto, CA, 94304,
USA
b
Department of Bioengineering, Stanford University, 443 Via Ortega, Stanford, CA, 94305, USA
c
Department of Radiology, Stanford University, 1201 Welch Road, Stanford, CA, 94305, USA

a r t i c l e i n f o a b s t r a c t

Keywords: International spread of the coronavirus SARS-CoV-2 has prompted many MRI scanning facilities to require scan
fMRI subjects to wear a facial covering (“mask”) during scanning as a precaution against transmission of the virus.
BOLD contrast Because wearing a mask mixes expired air with the subject’s inspired air stream, the concentration of inspired
Mask
carbon dioxide [CO2 ] is elevated, resulting in mild hypercapnia. Changes in the inspired gas mixture have been
Facial covering
demonstrated to alter R2∗ -weighted Blood Oxygen Dependent (BOLD) contrast. In this study, we investigate
COVID-19
SARS-CoV2 a potential for face masking to alter BOLD contrast during a sensory-motor task designed to activate visual,
auditory, and sensorimotor cortices in 8 subjects. We utilize a nasal cannula to supply air to the subject wearing
a surgical mask in on-off blocks of 90s to displace expired CO2 , while the subject performs the sensory-motor
task. While only a small fraction (2.5%) of the sensory-motor task activation is related to nasal air modulation,
a 30.0% change in gray matter BOLD signal baseline is found due to air modulation. Repeating the scan with
mask removed produces a small subject-specific bias in BOLD baseline signal from nasal air supply, which may be
due to cognitive influence of airflow or cannula-induced hypoxia. Measurements with capnography demonstrate
wearing a mask induces an average increase in ETCO2 of 7.4%. Altogether, these results demonstrate that wearing
a face mask during gradient-echo fMRI can alter BOLD baseline signal but minimally affects task activation.

1. Introduction
The widespread incidence of infection from the novel coronavirus
SARS-CoV-2 has prompted many research MRI scanning facilities to re-
quire continuous facial covering (wearing a mask) by scan subjects dur-
ing scanning in order to diminish the risk of virus transmission, espe-
cially since many of those infected are asymptomatic. The purpose is
primarily to contain respiratory droplets that can harbor the virus so
they avoid landing on surfaces or remain in the air potentially to infect
others. There are numerous styles of masks (CDC, 2020), ranging from
cloth and surgical procedure masks to N95 respirators, having various
degrees to which expired air is retained for one or more respiration cy-
cles. Wearing an efficacious mask will mix some expired air with fresh
air, and lead to increased carbon dioxide concentration [CO2 ] in in-
spired air. The question we approached is: Since CO2 is a potent va-
sodilator, does this elevation in [CO2 ] alter functional MRI (fMRI)?
fMRI is an epiphenomenological indicator of neural activity, rely-
ing on changes in blood oxygenation to depict metabolism changes
consequent to neuronal modulation, demonstrated in gradient-recalled
echo (GRE) imaging as changes in the susceptibility weighted relax-
ation rate R2∗ . The resulting Blood Oxygen Level Dependent (BOLD)
contrast (Ogawa et al., 1990) is subject to confounds that can alter the
oxygen content independent of neural metabolism including, for exam-
ple, changes in baseline cerebral blood flow (CBF) (Davis et al., 1998,
Kastrup et al., 1999, Buxton and Frank, 1997). Oxygen content of arte-
rial blood is affected by air exchange in the lungs; thus concentrations
of oxygen [O2 ] and carbon dioxide [CO2 ] in inspired air will be repre-
sented in blood chemistry delivered to the brain. In particular, breathing
air with increased [CO2 ] will cause increased arterial blood flow to vas-
cularized brain tissue, mainly in gray matter (Wise et al., 2007).
The purpose of this study is to examine the effect of wearing a mask
on BOLD contrast while subjects perform a robust sensory-motor (visual,
auditory, sensorimotor) task as the gas content in inspired air is manip-
ulated by periodically supplying fresh air through a nasal cannula. We
hypothesize that without a mask, supply of fresh air in the nasal cannula
will have minimal effect on air content and cause little change in BOLD
contrast due to CO2 (except for cognitive changes induced by the air
flow). On the other hand, any change in BOLD contrast due to fresh air
supply while wearing a mask is readily explained by the change in air
content (i.e. less CO2 accumulation during fresh air supply).
We first model the change in BOLD contrast that results from ele-
vated end tidal CO2 (ETCO2 ), and then describe fMRI scans to test the

∗
Corresponding author.
E-mail address: cslaw@stanford.edu (C.S.W. Law).

https://doi.org/10.1016/j.neuroimage.2021.117752
Received 22 October 2020; Received in revised form 16 December 2020; Accepted 4 January 2021
Available online 15 January 2021
1053-8119/© 2021 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
C.S.W. Law, P.S. Lan and G.H. Glover NeuroImage 229 (2021) 117752

Fig. 1. Simulation. (A) Model of BOLD contrast during task

with half-sinusoidal CMRO2 variation, square wave increases
in CBF during time frames 100-210, A = 0.02. Simulation set
maximum BOLD factor M = 1. (B) Sliding window analysis of
task contrast (Eq. 4b) does not vary substantially during ele-
vated CBF (except artifactually at transitions near frames 100
and 210). (C) Sliding window analysis of modeled task offset
(Eq. 4a) demonstrates baseline shift during elevated CBF. Δ𝐶
and Δ𝑆 are normalized by average task contrast.

degree and character of alterations that are observed while wearing a where 𝑑 is a constant which describes the amplitude of the task-elicited
mask. In separate measurements using capnography, we quantify the change and 𝐴(𝑡) is a square wave representing an additional positive
change in ETCO2 levels induced by wearing a mask. modulation of CBF due to air flow.
This model assumes that modulating [CO2 ] by airflow in the mask
2. Theory does not in itself cause BOLD changes from a potential cognitive in-
fluence of sensing airflow in the nose. Simulation results are shown
2.1. Simulation in Fig. 1A for T = 30 s, 270 time frames, 2 s sampling interval (TR),
A = 0.02, 𝑑 = 0.45, 𝑐 = 0.16 (Davis et al., 1998). The [CO2 ]-induced CBF
Hypercapnia causes an increase in BOLD signal (Wise et al., 2007), increase occurs during time frames 100 – 210 and leads to an elevated
due to elevated Cerebral Blood Flow (CBF), resulting from the va- BOLD baseline signal, but only an insignificantly decreased BOLD am-
sodilatory response to increased arterial blood [CO2 ]. The increase in plitude.
blood flow causes reduced deoxyhemoglobin concentration and a de-
crease in R2∗ . The relationship between BOLD signal (Δ𝑆 ), Cerebral 2.2. Quantification
Metabolic Rate of Oxygen (CMRO2), and CBF is given in the Davis model
(Davis et al., 1998) as a function of time (t) by In order to characterize the effect on BOLD contrast due to air manip-
{ ( )𝛽 ( )𝛼−𝛽 } ulation in the measured BOLD timeseries data, Δ𝑆(𝑡), a sliding window
Δ𝑆 (𝑡) CMRO2(𝑡) CBF(𝑡)
=𝑀 1− (1) analysis was performed to test for a baseline shift in the BOLD timeseries
Δ𝑆0 CMRO20 CB𝐹0
as well as for task contrast change. The baseline shift is
where M is the maximum BOLD contrast that can be obtained in 𝑡+nT∕2
1 ( )
the given brain region, 𝛼~0.4 is Grubb’s constant expressing an Δ𝑆 (𝑡) = Δ𝑆 𝑡′ dt ′ (4a)
empirically-derived power law relationship between blood volume and nT ∫𝑡−nT∕2
CBF (Grubb et al., 1974), 𝛽~1.5 is a constant relating signal change
The task design regressor is modeled by a sine wave with frequency
to concentration of deoxyhemoglobin determined by Davis, et. al, and
1/T. Therefore the sliding window task contrast change is approximated
the subscript 0 represents baseline values. We modeled the task-elicited
as
change in metabolism as
𝑡+nT∕2 ( )
1 ( ) 2𝜋𝑡′
CMRO2(𝑡) Δ𝐶 (𝑡) = Δ𝑆 𝑡′ sin dt ′ (4b)
= 1 + 𝑐𝐹 (𝑡) (2a) nT ∫𝑡−nT∕2 𝑇
CMRO20
where By setting window width to an integer multiple of the task period,
( ) () Δ𝑆 (𝑡) is therefore an estimate of slowly varying task signal amplitude.
⎧sin 2𝜋𝑡 , sin
2𝜋𝑡
>0
⎪ 𝑇 𝑇 Similarly, changes in Δ𝐶(𝑡) represent alterations in timeseries signal at
𝐹 (𝑡 ) = ⎨ ( ) (2b)
⎪ 2𝜋𝑡 the task frequency and can be taken to infer task contrast changes during
0, sin ≤0
⎩ 𝑇 the scan.
where 𝑐 is a constant, and T is the period of the task. Because the task de- This analysis was applied to a simulated timeseries with window du-
sign we used has an intentionally short block duration (15s), the evoked ration = 𝑇 (𝑛 = 1). We intentionally chose a small value for CBF change
hemodynamic response can be well modeled by a sine wave with fre- (A = 0.02) to mimic the small hypercapnic effects of wearing a mask.
quency 1/T. 𝐹 (𝑡) describes only positive increases in metabolism during The result is shown in Fig. 1B,C. As predicted, with elevated CBF hy-
on blocks. pothesized to occur due to increased ETCO2 with mask on and no added
Total change in CBF(𝑡) comprises the task-elicited change and the fresh air, there is a baseline shift Δ𝑆 (𝑡) but not a substantial task con-
change from turning air on: trast change Δ𝐶(𝑡) in BOLD signal. Note, however, if a subject’s breath-
CBF(𝑡) ing rate or mask efficacy changes during the scan, hypercapnia levels
= 1 + dF(𝑡) + 𝐴(𝑡) (3) could change. Consequent CBF variation would lead to added noise in
CBF0

2
C.S.W. Law, P.S. Lan and G.H. Glover
the BOLD signal, which could result in reduced statistical power in pre-
dicting task activation.
3. Methods
3.1. Experimental design
We sought to study whether wearing a mask would induce enough
hypercapnia from rebreathing expired air to affect BOLD activation
in a task. To activate visual, auditory, and sensorimotor regions, we
chose a robust block design sensory-sensorimotor (SM) task: the on-
block consisting of a contrast-reversing (flashing at 3Hz) checkerboard,
synchronously changing tones spanning one octave presented in pseudo-
random order, and subject sequentially tapping thumb to four digits on
the right hand synchronous in time with the tones and flashing checker-
board. The off-block is a fixation cross. The on and off blocks are inten-
tionally short (15s) so that the BOLD response is well represented by
a sine wave at the fundamental task frequency of 1/30Hz. During this
task, medical-grade air (air-on/off blocks; AIR) was supplied to the sub-
ject through a nasal cannula, in 90s duration on and off blocks (Fig. 2A).
The SM and AIR designs are nearly orthogonal to each other (correlation
coefficient = 0.013), which allows their characterization using a general
linear model (GLM).
A GLM with two design regressors is used to separately generate ac-
tivation maps of the sensory-motor (SM) task and hypercapnia (AIR)
effect. The sensory-motor task regressor is a square wave convolved
with hemodynamic response function (Glover, 1999), while the AIR re-
gressor uses an exponential response function (30s time constant sug-
gested during pilot scans) convolved with a square wave (Fig. 2B). The
9 minute-long scan was repeated with mask removed, which was ex-
pected to eliminate hypercapnic modulation of BOLD signal.
3.2. Subjects
Eight healthy subjects (3 males, 5 females, age 44 ± 16 years) from
Stanford University community were enrolled in the study after giv-
ing informed consent for a research protocol approved by the Stanford
University IRB. A nasal cannula (Hudson RCI flared, Teleflex Medical,
Research Triangle Park, NC) was placed in the subject’s nose and tubing
cinched comfortably around the back of the head. A pleated surgical
procedure mask with the metal nose strip removed (for RF and mag-
netic safety) was placed over the cannula and face, covering the nose
and extending to the chin. With the metal strip removed the mask did
3
NeuroImage 229 (2021) 117752
Fig. 2. (A) Sensory-motor (SM) and air-on/off (AIR) task
block design. (B) Regressors for SM and AIR. Scan duration is
540s long. Because of the short period of the SM task, its re-
gressor is approximated as a sine wave in subsequent sliding
window analyses.
not fit snugly to the face, so surgical tape was used to affix mask to face,
attempting to seal the gap. During the “mask-off’ scan the mask was slid
downward to uncover the nose and mouth. In this mode expired air was
not rebreathed because of the flow in the bore from the scanner’s bore
fan. The subject was able to perform the “mask-off” or “mask-on” ma-
neuver between scans, with minimal head motion. The order of mask-on
and mask-off scans was counterbalanced between subjects.
3.3. Breathing apparatus
Air was supplied to the cannula at a rate of 5.8 L/min from a tank of
medical-grade air through a regulator and solenoid valve controlled by
a computer also presenting the sensory-motor task stimuli with Eprime
software (Psychology Software Tools, Sharpsburg, PA). During both
“mask-on” and “mask-off” scanning, the airflow was cycled in 90s off-on
blocks (AIR) while subjects performed the SM task. The flow rate was
sufficient to provide fresh air within the mask approximately equal to
the average human breathing rate (Warner and Patel, 2018), displacing
expired air.
3.4. fMRI scanning
A 3T MRI scanner with 48-channel head coil (GE Premier, Milwau-
kee, WI) was employed. The scanner’s bore fan was turned on at low.
Anatomic scans were acquired with a 3D T1-weighted (BRAVO) se-
quence with 22cm FOV, 192 × 256 matrix, 32 4-mm thick slices, in
axial plane. Functional scans employed a gradient echo R2∗ -weighted
spiral-in/out pulse sequence (Glover and Law, 2001) with the same slice
prescription as the anatomic scan, 64 × 64 matrix, TR/TE 2000/30 ms,
77° flip angle, 270 time frames (9 minute scan duration). Physiological
data were acquired with the scanner’s respiration belt on the upper ab-
domen and pulse oximeter on the subject’s left index finger. Pneumatic
earphones and padding stabilized the head to diminish head motion.
No-mask fMRI data were not obtained for subject #8 (Table 1) due to
technical failure and inability to repeat the scan.
3.5. Activation maps
Postprocessing of timeseries images was performed with conven-
tional, homemade software, which included 6-parameter motion correc-
tion, correction of cardiac and respiration confounds using RETROICOR
(Glover et al., 2000) and RVHRCOR (Chang et al., 2009), and spatial
smoothing with a 1.5 pixel FWHM 3D Gaussian kernel. RETROICOR
C.S.W. Law, P.S. Lan and G.H. Glover NeuroImage 229 (2021) 117752

Table 1
Sliding window timeseries analyses.

Fractional task contrast change Δ𝐶𝑓 Fractional baseline shift Δ𝑆𝑓 Total RMS motion (mm)
Subject
With Mask No Mask With Mask No Mask With Mask No Mask

1 0.012 0.040 0.130 0.752 0.14 0.19

2 0.013 0.086 -0.028 0.110 0.18 0.12
3 -0.017 -0.034 0.032 -0.169 0.13 0.10
4 -0.024 0.033 0.103 0.119 0.12 0.11
5 0.134 0.082 1.726 0.042 0.18 0.14
6 0.078 -0.037 0.366 -0.381 0.12 0.17
7 0.000 -0.030 0.000 -0.017 0.27 0.16
8 0.006 — 0.071 — 0.20 —
Average 0.025 0.020 0.300 0.065 0.17 0.14

Table 1: Fractional change in task contrast Δ𝐶𝑓 and baseline shift Δ𝑆𝑓 for each subject. The AIR-related
average task contrast changes with or without mask (2.5%, 2.0%, respectively) were not significant
while the average baseline signal shift with mask-on (30.0%) was significant. The average baseline
signal shift of 6.5% (without mask) was not significant. No data were obtained on subject #8 without
mask due to technical failure. No significant difference in subject motion is noted between scans with
and without mask.

and RVHRCOR were not applied to subject #7 whose physiologi-
cal data were heavily correlated with the SM task. The GLM analy-
sis used quadratic detrending together with design regressors shown
in Fig. 2B. For generating group activation maps, FSL v6.0 software
(Jenkinson et al., 2012) was used to normalize activation maps to a
common MNI atlas.
3.6. Analysis
Timeseries were extracted from each scan using an ROI consisting of
those voxels with sensory-motor task activation T-scores exceeding 5.0,
(i.e, uncorrected p-values of 5e-7). The resulting timeseries was sub-
mitted to the sliding window analysis in Eq. 4(a,b) for scans with and
without mask for each subject.
To quantify the degree to which the time varying task contrast Δ𝐶(𝑡)
and baseline shift Δ𝑆 (𝑡) were affected by the airflow modulation, covari-
ation of these signals with the AIR regressor (shown in Fig. 2B) were cal-
culated for each subject’s mask-on and mask-off data. The results were
expressed as fractional changes with integrations covered the scan du-
ration:
∫ Δ𝑆 (𝑡) AIR(𝑡) dt
Δ𝑆𝑓 =
mean(Δ𝐶 (𝑡)) ∫ AIR(𝑡) dt
∫ Δ𝐶 (𝑡) AIR(𝑡) dt
Δ𝐶𝑓 =
mean(Δ𝐶 (𝑡)) ∫ AIR(𝑡) dt
3.7. Subject motion
Subject motion was estimated by calculating the brain centroid of
each volume before removing baseline drift or motion correction. Root-
mean-square (RMS) values of motion in x, y, and z dimensions were
computed. Total RMS motion is the square root of sum of squares of
RMS motions in x, y, and z.
3.8. ETCO2 measurement
End tidal CO2 measurements were made in separate sessions with
subjects outside the scanner, employing the identical or same type
of mask and cannula as utilized during scanning. The cannula was
connected to a patient monitor (Medrad Veris 8600, JZ Imaging,
Willoughby, OH) employing infrared gas capnography. Readings were
obtained every 15s for 5 min with mask in place as during scanning,
and with the mask removed. Two to three minutes of stabilization time
was employed before and between measurement blocks. The normalized
difference between mask-on and mask-off ETCO2 was computed as
( )
ΔETC𝑂2 = ETC𝑂2 on − ETC𝑂2 off ∕ETC𝑂2 off (6)
where ETC𝑂2 on,off denote the 5-minute-average readings with and with-
out the mask, respectively. A heteroscedastic T-test was employed to
determine the significance of the ΔETC𝑂2 in each subject, and a paired
T-test was employed to test the overall significance across the cohort.
To examine the possibility that breathing conditions could be different
in the magnet from the bore fan and confinement, the measurements
were repeated with subject #7 in the magnet performing the SM task
(but without scanning).
4. Results
4.1. Task activation
Fig. 3 shows single-subject activation results for the sensory-motor
task and for air-on/off blocks, each with and without mask. Task activa-
tion in Fig. 3A,B focused in visual, auditory, and sensorimotor cortices,
remains robust with or without mask. The difference between mask and
(5a) no mask task activation (maps not shown) is not significant. Supplying
air while wearing mask (Fig. 3C) reduces mask-induced hypercapnia,
resulting in widespread deactivation of BOLD signal predominantly in
gray matter, as expected. When the mask is removed (Fig. 3D), BOLD
(5b) signal related to air-on/off blocks demonstrates marked reduction, as
expected with normoxia. Deactivation was observed for the mask-off
condition in some individual subjects; for example, insula cortex and
frontal region in Fig. 3D. This suggests a cognitive influence derived
from sensation of airflow induced by the nasal cannula. With mask on,
some subjects reported a sense of nasal airflow. Some cognitive effect
of sensing nasal cannula airflow could be retained with the mask off.
Fig. 3E,F blue plots show timeseries of highly activated voxels, obtained
by averaging all voxels with SM activation having T scores > 5.0 (uncor-
rected p-value 5E-7), from scans shown in Fig. 3A,B. Also shown is the
AIR regressor, inverted to demonstrate that baseline signal is elevated
during air off periods with mask (Fig. 3E) but correlates negligibly with
AIR without mask (Fig. 3F).
The timeseries for each subject (as calculated for subject #3 in
Fig. 3E,F) was submitted to the sliding window analysis in Eq. 4(a,b).
Fig. 4 shows the result of averaging these sliding-window timeseries
across subjects to obtain the effect of air-on cycles on task contrast
change Δ𝐶 and baseline signal offset Δ𝑆 (𝑡).
Table 1 provides a summary of the sliding window analysis for each
subject presented as fractional changes in SM task contrast Δ𝐶𝑓 and

4
C.S.W. Law, P.S. Lan and G.H. Glover NeuroImage 229 (2021) 117752

Fig. 3. Single-subject (#3 in Table 1) activation results. (A), (B): Sensory-motor (SM) task (task > fixation). (C), (D): AIR effect (air-on > air-off). Color bars reflect
T-scores (red > 0, blue < 0). Deactivation in (C) demonstrates the baseline signal is larger with air-off due to mask-induced mild hypercapnia. (D) Without mask,
there is negligible global gray matter deactivation but some subjects exhibit cognitive deactivation due to sensing airflow. (E) Timeseries plot of activated voxels
(percentage change, T > 5.0, blue) from scan with mask (A), together with AIR regressor (red), which is plotted negatively to confirm signal is elevated during air
off. (F) Timeseries plot of same voxels with no mask scan in (B).

Fig. 4. Subject-averaged sliding window plots showing task contrast amplitude Δ𝐶(𝑡), baseline signal offset Δ𝑆 (𝑡), and AIR regressor (with sign reversed to demon-
strate its anticorrelation with Δ𝑆 ). Red bars on the figure represent Air-on periods while top blue bars represent SM task on-blocks. Signal was normalized by the
mean of task contrast for each subject before averaging.

5
C.S.W. Law, P.S. Lan and G.H. Glover NeuroImage 229 (2021) 117752

Fig. 5. Group activation results: (A)(B) Sensory-motor (SM) task (task > fixation). No significant group-level difference in SM task with and without wearing masks.
(C)(D) AIR effect (air-on > air-off). Global gray matter deactivation observed due to wearing masks (C) is not detected when masks were removed (D). Color bars
reflect T-scores. Mask results were averaged from 8 subjects while No Mask results were averaged from 7 subjects.

Table 2 when wearing a mask. A paired T-test across subjects confirmed that
End tidal CO2 measurements. this difference is significant at p < 0.0014. Our measurement of ETCO2
Subject ETCO2 , mm Hg increase from wearing a surgical mask is similar to that when wearing
ΔETCO2 (p value) N-95 masks (Bharatendu et al., 2020).
With mask No mask

1 36.8 33.1 0.111 (2E-14)

2 42.6 36.8 0.157 (3E-14) 5. Discussion and Conclusion
3 36.2 33.7 0.074 (5E-14)
4 35.6 32.9 0.080 (6E-06)
The experiments in this study employed a nasal cannula to replace
5 41.2 39.2 0.050 (1E-06)
6 31.2 30.6 0.021 (0.036) expired air with fresh air in a block design, thereby allowing controlled
7 36.0 33.7 0.066 (2E-08) manipulation of endogenous [CO2 ] levels during a sensory-motor task in
31.4∗ 29.5∗ 0.063∗ (2E-05∗ ) two separate scans (one with mask and one without mask). The results,
8 33.5 32.5 0.032 (5E-03)
while wearing a mask, demonstrated good qualitative agreement with
Average — — 0.074
the model; showing no significant change in task activation (p = 0.43)
Table 2: ETCO2 measurements with and without surgical mask. Normal- but significant change in baseline signal compared to the condition
ized fractional difference (ΔETCO2 ) was calculated from Eq.6. Subject when fresh air was introduced into nasal passages (p < 0.0141). These
#7∗ capnography data were collected in the scanner performing the SM results confirm that wearing a mask increases the BOLD baseline sig-
task but not being scanned. nal (30% on average, Table 1) through reduced R2∗ , but the effect does
not substantively alter gain of task-induced changes in signal; i.e, esti-
mates of task activation with and without mask (Table 1). Results with
fractional changes in baseline shift Δ𝑆𝑓 calculated by covariation with
no mask demonstrated negligible alteration to task activation, although
the AIR regressor (Eq. 5a,b) as well as subjects head motion information.
some subjects showed localized baseline signal shift correlated with air-
With mask on, the average baseline shift was 30.0% (p < 0.0141) while
on (Fig. 3D). This result was unexpected because there should not be a
the average task contrast change was not significant (2.50%, p = 0.43)
hypercapnic change with mask removed and air flowing. We speculate
from covariation with AIR. With mask removed, there was no signifi-
that such a baseline shift could be due to cognitive effects of airflow sen-
cant covariation of task contrast (2.0%, p = 0.45) or baseline shift (6.5%,
sation or a result of mild hypocapnia from fresh air being introduced in
p = 0.21) with AIR. Subject motion was small in this cohort of motivated
the nose. Nevertheless, Figs 3D and 5D illustrate that hypercapnic effect
volunteers, and not correlated with the AIR blocks. Paired-sample T-test
is largely absent with mask removed, as expected.
shows no significant difference in motion with and without mask.
The measurements of end tidal CO2 were made outside the scanner,
Sensory-motor task and air-on/off responses for single-subject maps
as quantitation was not possible while scanning and modulating the air
in Fig. 3 are as robust as responses averaged across subjects in Fig. 5.
using a cannula, because of mixing of the delivered air with the sam-
Sensory-motor task activation can be detected reliably under both mask-
pled expired air. Using a gas mask with rebreathing valve instead of
on and mask-off conditions (Fig. 5A,B) with no significant activation
nasal cannula would prevent the mixing effect and allow ETCO2 mea-
difference between conditions. Global gray matter deactivation is seen
surement while delivering air, but the substantial complexity would vi-
under mask-on and lack of airflow (Fig. 5C). Comparing to the entire
olate the concept of simply wearing a facial covering during fMRI. Re-
gray matter volume, the amount of deactivation (due to mild hypercap-
sults showed that reliable mild hypercapnia was induced with mask on
nic or cognitive effects from airflow) is insignificant without mask at
(average 7.4% increase in ETCO2 relative to no-mask condition). Differ-
group level (Fig. 5D).
ences with and without mask were significant within subjects (shown
in Table 2) and in the cohort (p < 0.0014, paired T-test). However, to
4.2. ETCO2 Measurements suggest that changes in ETCO2 measured in the control room were sim-
ilar to those during scanning, the measurements were repeated in the
Table 2 shows ETCO2 measurements on subjects outside the mag- scanner (without fMRI but performing the SM task) on one subject. The
net, as well as results on subject #7 in the magnet performing SM task capnography measurement on subject #7 inside and outside the scanner
but without scanning. The average change in ETCO2 was 7.4% increase were comparable as shown in Table 2.

6
C.S.W. Law, P.S. Lan and G.H. Glover
While there were individual differences in BOLD signal amplitude
and ETCO2 across subjects, we estimated the average BOLD reactivity
to ΔETC𝑂2 . For each subject, we calculated an individual AIR-activation
gray matter mask (mask-on data correlated with AIR regressor having
Z-score > 4 within gray matter). Each subject’s BOLD signal data was
filtered by their masks to calculate percent signal change. Averaging
each individual’s percent signal change per mmHg gave an estimate of
BOLD reactivity to ΔETC𝑂2 of 0.36%/mmHg. This value is slightly less
than that observed in gray matter by Prisman, et al. (Prisman et al.,
2008) (0.44%/mmHg) in 4 participants during controlled hypercapnia.
The difference may be due to lack of precision control of the mask and
cannula in our study. However, with limited number of subjects in both
studies, quantitation is underpowered.
While wearing a facial covering during fMRI does not substantially
alter task activation, it may create sensitivity to changes in respiration
or mask placement through variations in inspired [CO2 ] that are absent
without a mask. Some subjects may have experienced facial discomfort,
for example, prompting adjustments of their mask during scan. This ef-
fect would manifest as a difference between sliding-window timeseries
noise levels with and without mask. In the present study, no signifi-
cant differences were found in the task contrast change Δ𝐶(𝑡) between
scans with and without mask (Table 1), suggesting that wearing a mask
contributes a negligible effect to BOLD activation in our sensory-motor
experiments.
Our study is subject to a number of limitations: Only a small cohort
(8 subjects) could be recruited from within our lab because of institu-
tional regulations prompted by the pandemic. Of these, only 7 no-mask
scans were obtained for technical reasons. Only one type of mask was
evaluated. Different respirators may confine expired air more tightly
and exacerbate baseline signal change, although our results were sim-
ilar to observations made with N-95 masks (Bharatendu et al., 2020).
Only a robust sensory-motor task was evaluated, although effects on
verbal working memory were anecdotally tested and found to be con-
gruent with the SM results. In addition, we gathered three resting-state
data sets and analyzed the connectivity in the default mode network. Al-
though our initial results in functional connectivity, ALFF, FALFF, and
global signal show no significant effect from wearing a mask, a substan-
tially larger sample size would be needed to reliably detect small resting
state signal changes.
Because of recruitment limitations on the size of our subject cohort,
we used block-design air manipulations in an attempt to more robustly
estimate mask effects within a single scan than calculating differences
between separate scans with different airflow. In preliminary studies,
several subjects wearing a mask performed SM during separate scans,
twice each with and without continuous air supply (4 scans total). We
found that in the two repetitions without changing the air flow con-
dition, the SM task amplitude was not reproducible enough to make
reliable estimates of the difference of adding air to the cannula. Poor
reproducibility between scans, could presumably due to variations in
task performance, vigilance, and habituation, is consistent with earlier
studies of fMRI reproducibility (Voyvodic, 2006).
In summary, wearing a mask increases ETCO2 by 7.4% measured by
gas capnography. Increased [CO2 ] causes increased CBF and reduced
R2∗ , and induces global gray matter activation changes as in a CO2 gas
challenge (Wise et al., 2007) or breath holding (Kastrup et al., 1999).
This hypercapnic effect causes a change in signal baseline level but no
significant change in task activation.
NeuroImage 229 (2021) 117752
Funding
This work was supported by NIH grants R01 NS109450, P41
EB0015891.
Data availability
Data reported in this manuscript can be made available with permis-
sion from Stanford University.
Credit author statement
Christine S W Law: Conceptualization, Methodology, Formal analy-
sis, Investigation, Writing – Review & Editing, Visualization
Patricia S Lan: Conceptualization, Methodology, Investigation, Writ-
ing – Review & Editing
Gary H Glover: Conceptualization, Methodology, Formal analysis,
Investigation, Writing – Review & Editing, Visualization, Writing – Orig-
inal Draft, Project administration, Funding acquisition
Declaration of Competing Interest
The authors declare no conflict of interest.
References
CDC. Coronavirus Disease 2019 (COVID-19). Centers for Disease Control and Preven-
tion. https://www.cdc.gov/coronavirus/2019-ncov/prevent-getting-sick/about-face-
coverings.html. Published February 11, 2020. Accessed September 10, 2020.
Ogawa, S., Lee, T.M., Kay, A.R., Tank, D.W., 1990. Brain magnetic resonance imaging with
contrast dependent on blood oxygenation. Proc. Natl. Acad. Sci USA 87, 9868–9872.
Davis, T.L., Kwong, K.K., Weisskoff, R.M., Rosen, B.R., 1998. Calibrated functional MRI:
Mapping the dynamics of oxidative metabolism. Proc. Natl. Acad. Sci. USA 95,
1834–1839.
Kastrup, A., Li, T.Q., Glover, G.H., Moseley, M.E., 1999. Cerebral blood flow-related signal
changes during breath-holding. AJNR Am. J. Neuroradiol. 20, 1233–1238.
Buxton, R.B., Frank, L.R., 1997. A model for the coupling between cerebral blood flow
and oxygen metabolism during neural stimulation. J. Cereb. Blood Flow Metab. 17,
64–72. doi:10.1097/00004647-199701000-00009.
Wise, R.G., Pattinson, K.T.S., Bulte, D.P., et al., 2007. Dynamic forcing of end-tidal carbon
dioxide and oxygen applied to functional magnetic resonance imaging. J. Cereb. Blood
Flow Metab. 27, 1521–1532. doi:10.1038/sj.jcbfm.9600465.
Grubb, R.L., Raichle, M.E., Eichling, J.O., Ter-Pogossian, M.M., 1974. The effects of
changes in PaCO2 on cerebral blood volume, blood flow, and vascular mean tran-
sit time. Stroke 5, 630–639. doi:10.1161/01.str.5.5.630.
Glover, G.H., 1999. Deconvolution of impulse response in event-related BOLD fMRI. Neu-
roimage 9, 416–429. doi:10.1006/nimg.1998.0419.
Warner, M.T., Patel, B., 2018. Mechanical Ventilation. In: Hagberg and Benumof’s Airway
Management. Elsevier, Philadelphia, PA, pp. 804–820 e4.
Glover, G.H., Law, C.S., 2001. Spiral-in/out BOLD fMRI for increased SNR and reduced
susceptibility artifacts. Magn. Reson. Med. 46, 515–522.
Glover, G.H., Li, T.Q., Ress, D., 2000. Image-based method for retrospective correction of
physiological motion effects in fMRI: RETROICOR. Magn. Reson. Med. 44, 162–167.
Chang, C., Cunningham, J.P., Glover, G.H., 2009. Influence of heart rate
on the BOLD signal: the cardiac response function. NeuroImage 44.
doi:10.1016/j.neuroimage.2008.09.029, 857–69.
Jenkinson, M., Beckmann, C.F., Behrens, T.E., Woolrich, M.W., Smith, S.M., 2012. FSL.
NeuroImage 62. doi:10.1016/j.neuroimage.2011.09.015, 782–90.
Bharatendu, C., Ong, J.J.Y., Goh, Y., et al., 2020. Powered air purifying respirator
(PAPR) restores the N95 face mask induced cerebral hemodynamic alterations
among healthcare workers during COVID-19 Outbreak. J. Neurol. Sci. 417, 117078.
doi:10.1016/j.jns.2020.117078.
Prisman, E., Slessarev, M., Han, J., et al., 2008. Comparison of the effects of independently-
controlled end-tidal PCO2 and PO2 on blood oxygen level–dependent (BOLD) MRI. J.
Magn. Resonance Imaging 27, 185–191. doi:10.1002/jmri.21102.
Voyvodic, J.T., 2006. Activation mapping as a percentage of local excitation: fMRI stability
within scans, between scans and across field strengths. Magn Reson Imaging 24, 1249–
1261. doi:10.1016/j.mri.2006.04.020.